ADVANCED CMOS DEVICES AND RELIABILITY
BEN KACZER
UFRGS 2013 Tutorial
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
BTI
time
strain USJ
silicide
>130 90-65-45 Strain, USJ (F,C co-implant, …)
Performance
22-15
FinFET
Multi-gate 45-32/28
High-k, Metal Gate
FUSI
Gate-first
Gate-last
??
Ge/IIIV
III-V, CNT, VFET
TFET, NW, graphene…
nanowires
Tunnel FET
<15
VFET
SCALING ROADMAP : VARIETY OF
ARCHITECTURES ON THE HORIZON
EXPLOSION OF MATERIAL AND DEVICE TECHNOLOGY OPTIONS
Ge/III-V Strain
Metal gates FinFET
High k
HP -Logi c LP -Logi c Fl ash -FG
DRAM RRAM Fl ash -CT
POWER SUPPLY VOLTAGE EVOLUTION
0.00 1.00 2.00 3.00 4.00 5.00 6.00
0.01 0.10
1.00 10.00
Su pp ly v ol ta ge ( V)
Gate length ( µ m)
Power supply voltage is saturating at about 1V due to non-scaling of subthreshold slope
Constant voltage
Constant voltage
ELECTRIC FIELDS IN OXIDE AND IN SILICON INCREASE
Third scaling period: renewed constant voltage scaling due to non- scaling subthreshold slope has new implications on reliability !
1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07 1.0E+08
0.01 0.10
1.00 10.00
E lec tr ic field ( V /c m )
Gate length (um)
Eox Esi
70-80’s 90’s 00’s
POWER DENSITY INCREASES DUE TO NON-SCALING SUBTHRESHOLD SLOPE
Power density will soon be dominated by static power !
leak dd
stat
eff swing
dd dyn
I V P
: Static
f C
V V
P : Dynamic
⋅
=
⋅
⋅
⋅
⋅ α
=
Almost all reliability problems are accelerated by Temperature !
D. Cox, IRPS Tutorial 2004
NO TWO IDENTICALLY FABRICATED TRANSISTORS ARE ALIKE ANYMORE
These time-zero variations require adaptations in circuit design to account for statistical spread in device parameters
The established simulation
paradigm
Physical gate length 22nm
Physical gate length 9nm = 30x30x30
atoms
100
Asenov et al., IEDM 2008 σVT ~ 1/(WL)1/2
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
BTI
Reliability is an essential aspect of all products and technologies
IS RELIABILITY IMPORTANT ?
WHY RELIABILITY TESTING
As fabricated
Record performance!
Insufficient reliability: field returns, loss of profit, credibility, market share...
Devices cannot be tested for years (corresponding to field use)
accelerated testing needed!
V
DI
DE.g., a month later
Performance degraded!
V
DI
DACCELERATED TESTING (IN FEOL)
TYPICALLY DONE WITH V AND/OR T
Operating conditions Accelerated conditions
Log t im e- to -fa ilur e 10 years
Correct projection to operating conditions only possible if
same mechanism at accelerated and operating conditions
acceleration laws understood!
STOCHASTIC NATURE OF RELIABILITY MECHANISMS:
A FRACTION OF DEVICES WILL FAIL!
Operating conditions Accelerated conditions
Log t im e- to -fa ilur e
10 years
Again, the distribution can be correctly described only if underlying mechanism understood!
As dimensions get smaller,
distributions become wider
(courtesy of T. Grasser)
RELIABILITY MUST BE PART OF EVERY (TECHNOLOGY) SPECIFICATION!
Example:
1 out of 10,000 chips (100ppm) allowed to fail in
10 years at operating conditions.
“CLASSICAL” APPROACH TO RELIABILITY ASSESSMENT
failure data statistics
develop predictive reliability models accelerated
testing/model
reliability test structures failure
mechanisms
Reliability engineer Designer
Novel approaches needed to design reliable circuits with
unreliable components: Reliability Aware Design (RAD)
RELIABILITY AND RECENT CMOS FEOL TRENDS
• New (3D) device architectures
• New “exotic” materials (high-k gate dielectrics, metal gates, high-mobility substrates)
• Downscaling of devices toward atomic dimensions (variability)
• Supply voltages are not correspondingly reduced
Reexamination of known “old” degradation mechanisms in new materials and architectures required
Revision of existing and need for novel characterization techniques
New reliability lifetime assessment methodologies
Reliability cannot be guaranteed at technology level:
Reliability-Aware Design
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
BTI
TIME-DEPENDENT DIELECTRIC BREAKDOWN (TDDB)
Electrical stress = Additional gate oxide leakage paths (TDDB) + FET intrinsic parameters change (“BTI”)
G
S D
W
G
S D
W
G
S D
W
G
S D
W
G
S D
W
1 2
3 5
1 2 3 4 5
wear-out
4
10 -14 10 -12 10 -10 10 -8 10 -6 10 -4
5 4
3 2
1 0
Gate Voltage (V)
Fresh SILC
SBD HBD
CHANNEL HOT CARRIER (CHC) DEGRADATION
h
e
hole current
Electrons and/or holes in the channel of a MOSFET gain high energy under influence of large electrical fields in the Si at the drain. They are injected into the oxide causing interface traps to be created and charges to be trapped in the oxide
Bravaix et al., IRPS 2009
POSITIVE / NEGATIVE BIAS TEMPERATURE INSTABILITY (BTI)
0 5 10 15 20 25
0 500 1000 1500 2000
time (s) - ∆ V
th( m V)
T = 125
oC V
G,stress= -2 V V
G,relax= 0 V
stress relaxation 0V
0V V
G0V
Example: PFET at Negative gate Bias (and typically at elevated T emperature) pFET V
thstarts shifting (shows Instability) NBTI Charging of interface and oxide defects ∆V
thand ∆ μ
0.0 0.2 0.4 0.6 0.8 1.0
6 8 10 12 14 16 18
|V G-V th| [V]
Tinv (≈EOT+4Å) [Å]
EOT = 1nm
SiGe
Franco et al., IEDM 2010
ITRS
Critical issue in sub 1 nm oxides
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
• Overview / Phases of BD
• SILC
• SBD & impact of position
• Wear-out and HBD
BTI
V
Gate
Si substrate
(GIST OF) GATE OXIDE BREAKDOWN
I
t t
BD
D. R. Wolters, Ph.D. Thesis, 1985
40 nm SiO
2Gate oxide breakdown =
catastrophic failure
PHASES OF GATE DIELECTRIC BREAKDOWN
Electrical stress = Additional gate oxide leakage paths (TDDB) + FET intrinsic parameters change (“BTI”)
G
S D
W
G
S D
W
G
S D
W
G
S D
W
G
S D
W
1 2
3 5
1 2 3 4 5
wear-out
4
10 -14 10 -12 10 -10 10 -8 10 -6 10 -4
5 4
3 2
1 0
Gate Voltage (V)
Fresh SILC
SBD HBD
Crupiet al., TED 1998
SINGLE TRAP CONDUCTION PATH
Additional leakage path = Trap-assisted tunneling with energy loss
(Takagi et al., TED 46(2), 1999).This is Stress-Induced Leakage Current = SILC.
e
-V
Gn
+-type gate
p-type substrate
dielectric
1 0- 1 1
2 4
1 0- 1 0
2 4
1 0- 9
2 4
1 0- 8
1 01 5 1 01 6 1 01 7 1 01 8 1 01 9
p . Do t [ c m- 3]
Ielmini et al., TED 49 (11), 2002
• Steady-state SILC increase is
proportional to neutral oxide trap density
• Trap near the middle between interfaces
= highest conductivity
De Blauwe et al., TED 45(8), 1998
TWO COMPENSATING TRENDS IN ULTRA-THIN OXIDE SILC
Trend 1: trap generation rate drastically drops (figure)
▸ Less conductive paths through the oxide
Trend 2: conductivity per trap increases drastically
▸ Tunnel distance decreases and current rises exponentially
▸ Each trap contributes more to the total SILC
Stathiset al., IEDM 1998
T rap gener at ion ra te
THE CHANGING APPEARANCE OF SILC
SILC
time
many small contributions
oxide gate
substrate e-
SILC
time
few large contributions
oxide gate substrate e-
Capacitor top view
SILC and individual
trap components
Capacitor cross section
SILC
time
many small contributions
SILC
time
many small contributions
oxide gate
substrate e-
oxide gate
substrate e-
SILC
time
few large contributions
SILC
time
few large contributions
oxide gate substrate e-
oxide gate substrate e-
Capacitor top view
SILC and individual
trap components
Capacitor cross section
Thick / large device Thin / small device
Abrupt current increase of about 250nA
= creation of single trap conduction path
= ‘discrete’ SILC event
This is NOT a breakdown
TWO-TRAP CONDUCTION PATH
Percolation path with trap-assisted tunneling
▸ this is Anomalous Stress-Induced Leakage Current = a-SILC (non- volatile memories) (Okada et al., IEDM 2001, Degraeve et al., IEDM 2001) In thick oxide: moving bits in flash memory
▸ this is Micro-Breakdown (Cellere et al., TED 49(8), 2002) or pre- Breakdown (Degraeve et al., IEDM 2001)
2-trap path
capacitor top view e
-V
Gn
+-type gate
p-type substrate
dielectric
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
• Overview / Phases of BD
• SILC
• SBD & impact of position
• Wear-out and HBD
BTI
CONDUCTION PATH = SOFT BREAKDOWN
= SMALL GATE CURRENT INCREASE
• 1
stSBD observation (in 2 nm SiO
2):
Farmer et al., APL 52, p. 1749 (1988)
• Localized current flow through lowered barrier
• Model: ‘quantum point contact’
Suñé, IEDM 2000
• This is a soft breakdown, aka quasi-BD (
Lee, IEDM 1994), partial BD, B-mode
SILC (
Okada, VLSI 1997).
e
-V
Gn
+-type gate
p-type substrate
dielectric
<1 %-step detection mandatory in automatic tester or, use smaller areas
or, use a noise-based detection (
Roussel et al., TDMR 1(2), 2001)
11.4x10-6 11.2 11.0 10.8 10.6 10.4 10.2 10.0
gate current (A)
30 25
20 15
10 5
0
time (s) 2 percent
5 percent
soft breakdown
nmos transistor L x W = 1 x 100 µm2 Vg = 4.4 V
tox =2.4 nm
0 40 80 120
0 0.5 1.0 1.5
V
D(V) I
D( µ A)
G
S D
W
W
eff= 120 nm
V
G= 0, 0.15 .. 1.5 V
NO SIGNIFICANT EFFECT ON FET
CHARACTERISTICS AT THE MOMENT OF SBD
Kaczer et al., IRPS, p. 79 (2004)
Already in
Weir et al., IEDM, p. 73 (1997)
Circuits will generally work after SBD!
Gerrer et al., ESSCIRC 2009TIME-TO-SOFT BREAKDOWN =
STATISTICALLY DISTRIBUTED PARAMETER
Time-to-breakdown is Weibull distributed
Monomodal Weibull distribution is described as:
( )
⇒(
−(
−) )
= β( )
−β( )
η
− η
−
=
β
ln t
ln F
1 ln t ln
exp 1
t F
-4 -3 -2 -1 0 1
ln(-ln(1-F))
10
2 3 4 5 6 7
100
2 3 4 5 6 7
1000 tBD (s)
VG=4.4V
A=1.5e-08cm2 T=298C
EOT=2.4nm
η=63%-value β=slope
-5 -4 -3 -2 -1 0 1
ln(-ln(1-F))
0.01
2 4 6 8
0.1
2 4 6 8
1 normalized Q BD
4.6 11
tox= 2.4 nm 7.5
3.4
-5 -4 -3 -2 -1 0 1
ln(-ln(1-F))
0.01
2 4 6 8
0.1
2 4 6 8
1 normalized Q BD
4.6 11
tox= 2.4 nm 7.5
3.4
-5 -4 -3 -2 -1 0 1
ln(-ln(1-F))
-5 -4 -3 -2 -1 0 1
ln(-ln(1-F))
0.01
2 4 6 8
0.1
2 4 6 8
1 normalized Q BD
4.6
4.6 1111 7.5 tox= 2.4 nm 7.5
tox= 2.4 nm
3.4 3.4
dt t J QBD =
∫
0tBD G( )TDDB STATISTICAL DISTRIBUTION IS LINKED TO NUMBER OF TRAPS IN PERCOLATION PATH
Weibull slope increases with oxide thickness can be used to identify the breaking layer in a multilayer (high-k) stack
Not very precise or decisive for thin layers Very indirect monitor
14 12 10 8 6 4 2 0
Weibull slope
12 10
8 6
4 2
0
Physical layer thickness (nm) SiO2
CVS with VG > 0 V
Thick oxide
Thin oxide
DETERMINATION OF THE MAXIMUM APPLICABLE GATE VOLTAGE
Combined effect of temperature acceleration and statistical scaling (depends on β!) This extrapolation is only valid for soft breakdown
Voltage power-law more accurate than exponential also in high-k
Wu & Suñe, TED 2009
Soft BD
Degraeveet al., VLSI 1999
Temperature 0.6 V percentile
~0.3 V area
~0.25 V 10 years
G
S D
W
G
S D
W G
S D
W
0 L x
s is a monotonically increasing function of x.
R. Degraeve, IRPS (2001)
0 < s < 1
s ≅ 0 s ≅ 1
IMPACT OF BREAKDOWN POSITION ON FET
— PRINCIPLE OF POSITION DETERMINATION
D S
D
I I
s I
≡ +
Evaluate after BD in accumulation
Wu et al., IEDM, p. 187 (1998) Pompl et al., IRPS, p.82 (1999) Yang et al., IEDM, p.453 (1999)
EQUIVALENT ELECTRICAL CIRCUIT FOR FET AFTER BD
S D
W G
FS
Rpath
BS
FD BD
F
F
S, F
Dgate lengths, B
S, B
Dbase lengths a function of BD position
R
pathindependent of BD position G
S D
W
Explains varying FET behavior following BD
Kaczer et al., TED 49, p. 507 (2002)OTHER USES OF BD POSITION DETERMINATION
FinFETs:
Crupi et al., TED 53, p. 2351 (2006)Non-uniform oxide stress:
Crupi et al., EDL 24, p. 278 (2003)Processing issues:
Kauerauf et al., SISC (2004)Uniformity of GOX stress in accumulation and inversion:
Crupi et al., TDMR 3, p. 8 (2003)
Impact of BD position on BD wear-out:
Pey et al., IRPS, p. 221 (2007)Also
B. Linder et al., IRPS, p. 403 (2003)Correlation of successive BD’s and 1 and 2D:
Alam et al., p. 151, IEDM 2002; p. 415, IEDM2005;
TED 55, p. 3150 (2008)
These 2 cases are equivalent in 1D
x x
east south
west north
gate bulk
1 μm
Kaczeret al., IPFA 2007
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
• Overview / Phases of BD
• SILC
• SBD & impact of position
• Wear-out and HBD
BTI
CONDUCTION PATH + THERMAL DAMAGE
= HARD BREAKDOWN
Mass transport and Si regrowth in the breakdown spot
Lei Jun Tang et al., TDMR 4(1), 2004
Localized current flow through resistor (+diode)
Kaczer et al., TED 49(3) 2002
V
Gn
+-type gate
p-type substrate
dielectric
capacitor top view
hard breakdown
PHASES OF DEGRADATION AND WEAR OUT IN THIN OXIDES
1. SBD: Weibull ( β
SBD, η
SBD)
2. Wearout: assume Weibull ( β
wo, η
wo)
3. Runaway
HBD
R
S-limited
DETAILED STRUCTURE OF THE WEAR OUT PHASE
In some cases the wear out phase is clearly observable by the increased ‘digital’ noise (Digital Breakdown)
J.S. Suehle et al., IRPS 2004
Kaczer et al., IEDM 2004
0 500 1000
100 101 102 103 104
t
rs(s)
gate current (A)
• Analog breakdown rate increases with lower R
s(e.g. in Metal gates)
• Can be “arrested” by limited current supply (e.g. from previous stage)
ln(10)
( ) log( ) g ... log( )rs ( g)
s s
I t t V C t D V
a R R
= + + ≡ +
Alam et al., IEDM 2000; Linder et al., VLSI 2001; Monsieur et al., IRPS 2003;
Kaczer et al., IRPS 2004
HARDNESS OF BREAKDOWN
DEPENDS ON GATE ELECTRODE
Post-BD current through BD path depends on gate material Gate current run-away is positive feed back process
controlled by electron ‘supply’ in the gate
metal gate hard BD
poly-Si gate soft BD
Duschl et al., IRPS, p. 632 (2005).
-3 -2 -1 0 1
ln (-l n (1 -F ))
1
2 4 6
10
2 4 6
100
2 4 6
1000 time-to-breakdown (s)
V
G=2.2V
0.9 nm EOT SiO
2/HfO
2A=1.25e-09cm
2HBD β
HBD=~1.2 SBD β
SBD=~0.7
HBD IS NOT WEIBULL DISTRIBUTED!
If HBD is fitted with Weibull, β will be too high erroneous conclusions can be drawn
TDDB data meaningful if BD detection criterion given
t With limited data, hard breakdown resembles I
Weibull distribution
Degraeve et al., IRPS 2006 Kerber et al., IRPS 2007
virgin device 1st SBD
COMPETING FOR HBD: A TRIATHLON …
In statistics described by series rule of reliability
2nd SBD 3rd SBD
1st HBD
faster swimmer can be overtaken by even faster biker!
time
SBD HBD
ST AR T
Winnercompetitor 1
competitor 2
P. J. Roussel et al., Microelectron. Eng. 84, 1925 (2007)
ALL-IN-ONE RELIABILITY PREDICTION
Region 1 = breakdown-free
Region 2 = multiple soft breakdown
▸ Contours = relative leakage current increase DI/I
0after 10 years
▸ NOTE: current increase for individual small transistor can be substantial!
Region 3 = Hard breakdown (taking multiple SBD into account)
▸ Contours = lifetime for 0.01% failures
0.001
2 4 6
0.01
2 4 6
0.1
Area (cm2 )
1.4 1.3
1.2 1.1
1.0
Voltage (V) 10
8.4 77
s =1 0 y
ear s
10 s 8
10 s 7
10 s 6
10 s5
10 s 4
Region 3 Region 1
Region 2
3x10-6 10-43x10-5
10-5
0.001
2 4 6
0.01
2 4 6
0.1
Area (cm2 )
1.4 1.3
1.2 1.1
1.0
Voltage (V) 10
8.4 77
s =1 0 y
ear s
10 s 8
10 s 7
10 s 6
10 s5
10 s 4
Region 3 Region 1
Region 2
3x10-6 10-43x10-5
10-5
SiO
2/HfO
2EOT=0.9 nm
Sahhaf et al., IEDM 2007
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
BTI
• Overview / BTI components
• Technological remedies
• NBTI recoverable component and AC stress
• BTI in deeply scaled devices
• Observing properties of individual defects
• Impact of individual defects on FET characteristics
• Constructing and projecting ∆Vth distributions
• Combining time-dependent and time-0 variabilities
• Defect-centric circuit simulation framework
BIAS AND TEMPERATURE STRESS:
TYPICAL DURING FET OPERATION
V
DD0V V
DD0 5 10 15 20 25
0 500 1000 1500 2000
time (s) - ∆ V
th( mV )
T = 125
oC V
G,stress= -2 V
0V
0V negative V
G0V
Example: PFET V
that Negative gate Bias (and typically at elevated T emperature) starts shifting (shows Instability) NBTI
Charging of interface and oxide defects ∆V
thand ∆ μ
NBTI (pFET) PBTI (nFET)
Positive oxide charge (negative V
thshift)
Negative oxide charge (positive V
thshift) Interface states
(mobility, transconductance, subthreshold-slope degradation)
Mahapatra, IEDM 2003 Mitani et al., ECS 2005 Garros et al., ICICDT 2010
X
(but observed in sub 1nm EOT)
Generally at substrate interface Generally trapping in high-k
BTI: CHARGED OXIDE BULK AND INTERFACE STATES
Zafar et al., VLSI 2006 Mahapatra, TDMR 2007
Kerber et al., TED 55, 3175 (2008) Aoulaiche et al., IRPS 2009
EXAMPLE OF BTI LIFETIME EXTRAPOLATION
▸ Failure Criterion: ∆ V
th= 30mV at T = 125
◦C
▸ Minimized relaxation: t
sense_delay= 2ms
▸ Stress experiment at different voltages (V
Gstress> V
DD) Estimate lifetime Power Law extrapolation at 10 Years
VGstress-Vth0=-1.25V
BTI IS A SERIOUS RELIABILITY PROBLEM
Ultra-Thin EOT high-k gate stacks with overdrive below the required 0.7V for 10 years
nMOS - PBTI pMOS - NBTI
1.4 1.2 1.0 0.8 0.6 0.4 0.2 PBTI Vg overdrive at 10years (V) 0.0
1.6 1.2
0.8 0.4
0.0
EOT (nm) Hf-based
GdHfO Zr-based
target: 0.7V
1.4 1.2 1.0 0.8 0.6 0.4 0.2 NBTI Vg overdrive at 10years (V) 0.0
1.6 1.2
0.8 0.4
0.0
EOT (nm) target: 0.7V
Si substrate
Courtesy of Cho and Franco, imec
“MINOR” COMPLICATION: THRESHOLD VOLTAGE SHIFT STARTS RELAXING IMMEDIATELY AFTER STRESS REMOVED
0 5 10 15 20 25
0 500 1000 1500 2000
time (s)
∆ V
th( m V)
T = 125
oC V
G,stress= -2 V V
G,relax= 0 V stress relaxation
Crucial issue for all BTI measurements and interpretation!
Observed in p/nFETs, P/NBTI, DC and AC stress...
Alam, IEDM 2003; Kerber et al., TED 2008; Kaczer et al., IRPS 2008
BTI DEGRADATION RELAXES “DISPERSIVELY” OVER MANY TIME SCALES: µSECONDS OR LESS TO DAYS
10
-310
-110
110
310
5relaxation time t
relax(s)
“DC” MSM Fast MSM
Ultra-Fast MSM On-the-fly
Extended MSM
Different measurement techniques to attack the problem
MSM = Measure- Stress- Measure
10
-5degr adat ion ∆ V
thRELAXATION INCOMPLETE:
PERMANENT DEGRADATION
0 10 20 30 40 50
10
-310
-110
110
310
5relaxation time t
relax(s)
∆ V
th( m V)
“Permanent”
R(t
stress, t
relax) R (
t stress,t
relax= 0)
1.4 nm SiON pFET t
stress= 100 s T = 125
oC
Most common interpretation:
Relaxation component – discharging of oxide defects Permanent component – generated interface states
se e
Ranganet al., IRPS 2003OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
BTI
• Overview / BTI components
• Technological remedies
• NBTI recoverable component and AC stress
• BTI in deeply scaled devices
• Observing properties of individual defects
• Impact of individual defects on FET characteristics
• Constructing and projecting ∆Vth distributions
• Combining time-dependent and time-0 variabilities
• Defect-centric circuit simulation framework
PBTI CONTROLLED BY CHARGE TRAPPING IN HIGH-K
10
-410
-310
-210
-110
01
E
ox(MV cm
-1) Δ V
th( V)
2 5 10
ref La
ref La
T = 125 oC trelax = 10-3 s lines: tstress = 10 years symbols: tstress = 6x103 s
• Group III elements compensate unpaired electrons around oxygen vacancy in HfO
2• Similar effect caused by N (
Xiong & Robertson, JAP 2006; Sahhaf et al., EDL 2010)
Kaczeret al., INFOS 2009
Reference
With La
Liu and Robertson, APL 2009
Significant reduction of PBTI is observed in nFETs with Lanthanum V
Th0-adjusting layer
SIGE PFETS WITH BURIED CHANNEL PROMISE SUPERIOR PERFORMANCE OVER SI
∆EV
Si SiGe Si SiO HfO MG
To reduce NBTI:
• Increase Ge fraction
• Increase QW thickness
• Reduce Si cap thickness
Franco et al., IRPS 2010, IEDM 2010Mitardet al., IEDM 2010
64
0 0.2 0.4 0.6 0.8 1 1.2 1.4
6 8 10 12 14 16 18
Max.
|V
G-V
th|
for 10Y[V]
T
inv[Å]
Ultra-Thin EOT
0 1 2 3 4 5 6 7 8 9
6 8 10 12 14 16 18
Max.
E
oxfor 10Y[M V/ c m ]
T
inv[Å]
THIN Si CAP IMPROVES NBTI LIFETIME WHILE SCALING T
INV▸ Enables T
invscaling while improving NBTI reliability
▸ Same trend consistently observed on SiGe, Ultra-Thin EOT SiGe, and pure Ge channels
Si cap thinner
thicker
0.65nm 1nm
1.3nm
2nm
Si cap thinner
thicker
0.65nm
1.3nm 2nm
SiGe
1nmSi
0.6nm 1nm
1.7nm 0.6nm
1nm
1.7nm
J. Franco et al. IEDM 2010
64
Thin Si cap: channel carriers (E
F) ‘miss’ part of the oxide defects
SiGe LOWER ∆N
OTMODEL: REDUCED INTERACTION WITH OXIDE DEFECTS
SiGe Si SiO2 HfO2 Si
E
FE
VSMG
▸ Less defects, located on the gate side, are energetically favorable for holes
▸ Stronger field acceleration due to ‘unaccessible’ traps becoming ‘accessible’
R ↓
J. Franco et al. IEDM 2011
Thick Si cap: channel carriers (E
F) ‘see’ all the oxide defects
E
FE
VSSiGe Si SiO2 HfO2
Si MG
SiGe LOWER ∆N
OTMODEL: REDUCED INTERACTION WITH OXIDE DEFECTS
▸ Less defects, located on the gate side, are energetically favorable for holes
▸ Stronger field acceleration due to ‘unaccessible’ traps becoming ‘accessible’
R ↓
J. Franco et al. IEDM 2011
UNIFIED PICTURE: THE ENERGY ALIGNMENT BETWEEN CARRIER AND OXIDE DEFECTS
CONTROLS BTI MECHANISMS
▸ nMOSFET PBTI: improved reliability shifting up the defect energy level by inserting Gd (Dy, La) in the high-k
▸ pMOSFET NBTI: improved reliability shifting up the
channel Fermi level by using a (Si)Ge channel
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
BTI
• Overview / BTI components
• Technological remedies
• NBTI recoverable component and AC stress
• BTI in deeply scaled devices
• Observing properties of individual defects
• Impact of individual defects on FET characteristics
• Constructing and projecting ∆Vth distributions
• Combining time-dependent and time-0 variabilities
• Defect-centric circuit simulation framework
COMMON PROPERTY OF NBTI RELAXATION (RATE) AND LOW-F (“1/F”) NOISE:
STATES WITH WIDELY DISTRIBUTED TIME SCALES
t
stress= 100 s V
stress= -2V T = 125
oC
0 10 20 30 40 50 60 70
10-3 10-1 101 103 105
t
relax(s) Δ V
th(m V)
10-8 10-6 10-4 10-2 100
-d Δ V
th/ d t (Vs
-1)
10-16 10-15 10-14 10-13 10-12 10-11 10-10 10-9
100 101 102 103 104 105
frequency (Hz) S
VG(V
2Hz
-1)
SiON T = 25 oC unstressed
VG = -0.3 V
VG = -0.6 V
NBTI relaxation & rate low-f noise (same device)
Both 1/f noise and R component higher in nitrided oxides
Kapila et al., IEDM 2008; Kaczer et al., IRPS 2009
… …
LEARNING FROM NOISE: TWO ATTRIBUTES ESSENTIAL FOR CAPTURING BTI RELAXATION BEHAVIOR
C = 10
0... 10
NC = 10
0... 10
Nlow-f noise
(1) Many time scales distributed RC components
NBTI relaxation
(1) + (2) Different charging (capture) and discharging (emission) of each state
Kaczeret al., IRPS 2008, 2009
capture time constant (s)
emission time constant (s) Reisinger et al., IRPS 2010
PFET AC NBTI: (FAST) SEQUENCE OF STRESS AND RELAX PHASES
0 10 20 30 40 50
10
010
210
410
610
810
10frequency (Hz) Δ V
th( m V)
DC
T = 125 C
t
stress= 1000 s
V
S= V
D= V
B= V
cc= 2 V
AC NBTI:
V
select= 2 V DC NBTI:
V
select= 0 V V
G= 0 V
AC degradation ~ DC degradation / 2.
• AC BTI degradation lower than DC
• appears f-independent up to GHz
|VG|
t stress relax stress relax
…
Zhu et al., IRW 2002 Abadeer, IRPS 2003 Alam, IEDM 2003 Chakravarthi et al., IRPS 2004 Fernandez et al., IEDM 2006
0 20 40 60 80 100
0 20 40 60 80 100
Duty Factor
Δ V
th= R (t
relax) + P (m V)
Vstress = -2.0V fstress = 10 kHz tstress = 6000 s
trelax = 10-3 ( ), 10-2, … , 104 s T = 125 oC
relaxation
DISTINCTIVE SHAPE OF AC NBTI Δ V
THDEPENDENCE ON STRESS DUTY FACTOR
• Shape with “plateau” due to recoverable component
• Correct model must reproduce this shape
DC
DUTY FACTOR DEPENDENCE OF THE RECOVERABLE COMPONENT REPRODUCED BY EQUIVALENT CIRCUIT
Kaczer et al., IRPS 2008
0 1 2 3 4 5 6 7
0 20 40 60 80 100
Duty Factor R(t relax, t stress= 6000) (a.u.)
Vstress = 1 a.u.
trelax = 10-3, 10-2, … , 104a.u.
relaxation
Experiment Equivalent circuit simulation
• Distinctive shape of Duty Factor dependence cannot be explained by the classical Reaction-Diffusion model
• The shape is controlled by the recoverable component
0 20 40 60 80 100
0 20 40 60 80 100
Duty Factor ΔVth= R(trelax) + P (mV)
Vstress= -2.0V fstress= 10 kHz tstress= 6000 s
trelax= 10-3 ( ), 10-2, … , 104s T= 125 oC
relaxation
DC
OUTLINE
CMOS FEOL trends Why reliability
Examples of FET degradation mechanisms TDDB
BTI
• Overview / BTI components
• Technological remedies
• NBTI recoverable component and AC stress
• BTI in deeply scaled devices
• Observing properties of individual defects
• Impact of individual defects on FET characteristics
• Constructing and projecting ∆Vth distributions
• Combining time-dependent and time-0 variabilities
• Defect-centric circuit simulation framework
DEEPLY-SCALED DEVICE OPERATION IS AFFECTED BY INDIVIDUAL DEFECTS
In deeply-downscaled technologies, only a handful of random defects will be present in each device
N
ot= 10
12cm
-2 N
T~ 10 if device area = 10 x 100 nm
2Courtesy of M. Bina, TUWien
Number of charged defects will be increasing with operating time
time-dependent variability in addition to time-0 variability
BEN KACZER, IMEC BEN KACZER, IMEC
INDIVIDUAL DEFECTS RESULT IN TIME-DEPENDENT VARIABILITY
78
• Individual defects behave stochastically
• Individual defects have considerable relative impact on device
• These time-dependent variations require adaptations in circuit design to account for time-dependent statistical distributions of device
parameters
Large devices
Large Transistor 280 x 720 nm2
N
T= 800 traps
Lifetime
Grasser et al., IEDM 2010
(c)
lifetime (s)
Occurrence
Lifetime
Small device 35 x 90 nm2
N
T= 12 traps
Lifetime
DEEPLY SCALED DEVICES: INDIVIDUAL EMISSION EVENTS VISIBLE IN NBTI RELAXATION TRACES
relaxation time -Δ V
th( m V)
10-3 10-1 101 103 105
Large pFET
[Asenov]
0 10 20 30 40 50 60
relaxation time (s) -Δ V
th( m V)
10-3 10-1 101
70 × 90 nm
2pFET
…
τ
n-1τ
1τ
2τ
n-2τ
nKaczeret al., IRPS 2009
IN GENERAL, EACH DEFECT CHARACTERIZED BY
• capture time τ
c• emission time τ
e• impact on device (∆I
dor ∆V
th,...)
• occupancy (0 or 1) at given time
(covers BTI, RTN, ... easily extensible to other mechanisms)
(size of bubble represents impact)
Example: defects in 3 different devices
τ
e(V, T) τ
c( V , T )
τ
e(V, T) τ
c( V , T )
τ
e(V, T) τ
c( V , T )
EACH DEVICE CHARACTERIZED BY
• number of defects N
Twith above properties
Depend on
• spatial position
• energy position
• lattice relax. energy
• ...
Kaczer et al., IRPS 2011
EXTENDED MEASURE-STRESS-MEASURE (eMSM):
EFFICIENT WAY OF OBSERVING MOST DEFECTS
~ log stress time degr adat ion (~ to ta l Δ V
th)
Example: 3 deeply-scaled devices
~ log relaxation time degr adat ion (~ to ta l Δ V
th)
τ
e(V
L, T) τ
c( V
H, T )
τ
e(V
L, T) τ
c( V
H, T )
τ
e(V
L, T)
τ
c( V
H, T ) t
stresst
relaxNBTI AND RTN: TWO SIDES OF THE SAME COIN
Random Telegraph Noise (steady-state)
NBTI relaxation transient (perturbation/non-steady-state)
S
+ + + + + +D
+ + +
+
0 10 20 30 40
10-3 10-2 10-1 100
-Δ V
th(mV )
time (s) t
stress= 1900 s
101 102 103
+ + + + +
+ + +
S D
-15 -10 -5 0 5 10 15
0 100 200 300 400
time (s) t
stress= 0 s
Δ V
th(mV )
B. Kaczer et al., IRPS 2009 & 2010